Growing Economies from the Bottom Up. Presenter: Leigh Tesfatsion

Transcription

1 Agent-Based Computational Economics Growing Economies from the Bottom Up Presenter: Leigh Tesfatsion Professor of Economics and Mathematics Department of Economics Iowa State University Ames, Iowa

2 Outline What is Agent-based Computational Economics (ACE) in a nutshell? Simple labor market illustration (implemented via the TNG Lab) Four strands of current ACE research Potential advantages and disadvantages of ACE for economic modeling 2

3 What is ACE? Computational study of economic processes as dynamic systems of interacting agents A culture-dish approach to the theoretical study of economic processes 3

4 ACE Culture-Dish Analogy Modeler constructs a virtual economic world populated by various agent types Modeler sets initial world conditions Modeler then steps back to observe how the world develops over time without intervention (no imposed equilibrium, rational expectations, etc.) World events are driven by agent interactions 4

5 ACE Agent Types Agents = Encapsulated software programs representing individual, social, biological and/or physical entities Cognitive agents are capable (in various degrees) of Behavioral adaptation Social communication Goal-directed learning Endogenous evolution of interaction networks Autonomy Autonomy (self-activation and self-determinism based on private internal processes) 5

6 Initial World Conditions (Experimental Treatment Factors) Structural conditions Institutional arrangements Behavioral dispositions of agents 6

7 ACE Culture Dish Analogy Initial World Conditions (Experimental Treatment Factors) World Develops Over Time (Culture Dish of Agents) Macro Regularities 7

8 Illustrative ACE Application Area: Labor Institutions and Market Performance Some Key Issues: Labor contracts typically incomplete Supplemented by government programs with numerous eligibility restrictions Difficult to test program effects by means of conventional analytical and/or statistical tools 8

9 Example: U.S. State Programs Providing Example: Unemployment Benefits (UB) Typical Features of State Programs (e.g., Iowa): UB only paid to no fault of their own unemployed UB recipients must continue to seek employment UB levels based on past earnings UB of limited duration UB financed by employer contributions at rates determined in part by each employer s benefit ratio = [UB paid out to former employees divided by the employer s taxable payroll] Additional UB often granted when unemployment rate is abnormally high for prolonged periods Complicated Rules!! 9

10 ACE Labor Market UB Study Pingle/Tesfatsion 2003 (Experiments Implemented via TNG Lab) W1 W2 W3... W12 E1 E2 E3... E12 Preferential job search (workers W employers E) with choice/refusal of partners: Purple directed arrow = Refused work offer. 10

11 ACE Labor Market 12 workers with same observable structural attributes in initial period T=0 12 employers with same observable structural attributes in initial period T=0 Only observable source of heterogeneity among workers and among employers is their expressed behaviors on the work-site 11

12 ACE Labor Market Each worker can work for at most one employer in each period T Each employer can provide at most one job opening in each period T Work-site strategies in initial period T=0 are randomly determined and private information 12

13 Each worker and employer has Publicly available information about various market/policy protocols (e.g., unemployment benefit eligibility rules) Private behavioral methods that can evolve over time Privately stored data that can change over time 13

14 A Computational Worker Public Access: // Public Methods Protocols governing job search Protocols governing negotiations with potential employers Protocols governing unemployment benefits program Methods for receiving data Methods for retrieving Worker data Private Access: // Private Methods Method for calculating my expected utility assessments Method for calculating my actual utility outcomes Method for updating my worksite strategy (learning) // Private Data Data about myself (my history, utility fct., current wealth ) Data recorded about external world (employer behaviors, ) Addresses for potential employers (permits communication) 14

15 A Computational Employer Public Access: // Public Methods Protocols governing search for workers Protocols governing negotiations with potential workers Protocols governing unemployment benefits program Methods for receiving data Methods for retrieving Employer data Private Access Only: // Private Methods Method for calculating my expected profit assessments Method for calculating my actual profit outcomes Method for updating my work-site strategy (learning) // Private Data Data about myself (my history, profit fct., current wealth ) Data recorded about external world (worker behaviors, ) Addresses for potential workers (permits communication) 15

16 Flow of Activities in the ACE Labor Market Workers make offers to preferred employers at a small cost per offer (quits allowed) Employers accept or refuse received work offers (firings allowed) Each matched pair engages in one work-site interaction (PD game - cooperate or defect) Any unemployed (unmatched) worker or vacant (unmatched) employer receives a UB payment After 150 work periods, each worker and employer updates its work-site strategy 16

17 Flow of Activities in the ACE Labor Market Initialization Do 1000 Loops Work Period: Search/Match Worksite Interactions Update Expectations Do 150 Loops Evolution Step: Evolve Worksite Strategies 17

18 Worksite Interactions as Prisoner s s Dilemma (PD) Games Employer C D C (40,40) (10,60) Worker D (60,10) (20,20) D = Defect (Shirk); C = Cooperate (Fulfill Obligations) 18

19 Key Issues Addressed How do changes in the level of the unemployment benefits (UB) payment affect... Worker-Employer Interaction Networks Worksite Behaviors: Degree to which workers/employers shirk (defect) or fulfill obligations (cooperate) on the worksite Market Efficiency (total surplus net of UB program costs, unemployment/vacancy rates,...) Market Power (distribution of total net surplus) 19

20 Experimental Design Treatment Factor: Unemployment Benefits Payment (UB) Three Tested Treatment Levels: UB=0, UB=15, UB=30 Runs per Treatment: 20 (1 Run = 1000 Generations; 1 Gen.=150 Work Periods) Data Collected Per Run: Network patterns, behaviors, and market performance (reported in detail for generations 12, 50, 1000) 20

21 Three UB Treatments in Relation to PD Payoffs UB=0 < L=10 L=10 < UB=15 < D=20 D=20 < UB=30 < C=40 NOTE: Work-site PD payoffs given by: L (Sucker)=10 < D (Mutual-D)=20 < C (Mutual-C)=40 < H (Temptation)=60 21

22 Market Efficiency Findings As UB level increases from 0 to 30 higher average unemployment and vacancy rates are observed; KNOWN EFFECT more work-site cooperation observed on average among workers & employers who match. NEW EX POST EFFECT Note: These outcomes have potentially offsetting effects on market efficiency. 22

23 Efficiency Findings... Market Efficiency (Utility less UB Program Costs) Averaged Across Generations 12, 50, and 1000 for three different UB treatments Market Efficiency UB 23

24 Efficiency Findings... UB=15 yields highest efficiency UB=0 yields lower efficiency (too much shirking) UB=30 yields lowest efficiency (UB program costs too high) 24

25 Multiple Attractors Two distinct attractors observed for each NEP treatment... UB=0 and UB=15: First Attractor = Latched network supporting mutual cooperation; Second Attractor = Latched network supporting intermittent defection UB=30: First Attractor = Latched network supporting mutual cooperation Second Attractor = Completely disconnected network (total coordination failure) 25

26 Multiple Network Attractors Two distinct attractors observed for each UB treatment... No UB (0) or Low UB (15) : First Attractor = Latched W-E network supporting mutual cooperation; Second Attractor = Latched W-E network supporting intermittent defection High UB (30): First Attractor = Latched network supporting mutual cooperation Second Attractor = Completely disconnected network (total coordination failure) 26

27 The Following Diagrams Report... 1 Two-sided (W-E) network distributions 0=Stochastic fully connected network 12=Latched in pairs W W... E E 24=Completely disconnected 2 Worksite behaviors supported by these network outcomes 27

28 Network Distribution for UB=0 Sampled at End of Generation 12 Network Distribution for ZeroT: Number of Runs Network Distance Intermittent Defection Mutual Cooperation 28

29 Network Distribution for UB=0 Sampled at End of Generation Network Distribution for ZeroT: Number of Runs Network Distance Intermittent Defection Mutual Cooperation 29

30 Network Distribution for UB=0 Sampled at End of Generation Network Distribution for ZeroT: Number of Runs Network Distance Intermittant Defection Mutual Cooperation 30

31 Network Distribution for UB=15 Sampled at End of Generation Network Distribution for LowT: Number of Runs Network Distance Intermittent Defection Mutual Cooperation 31

32 Network Distribution for UB=15 Sampled at End of Generation Network Distribution for LowT: Number of Runs Network Distance Intermittent Defection Mutual Cooperation 32

33 Network Distribution for UB=15 Sampled at End of Generation Network Distribution for LowT: Number of Runs Network Distance Intermittent Defection Mutual Cooperation 33

34 Network Distribution for UB=30 Sampled at End of Generation Network Distribution for HighT: Number of Runs Network Distance Intermittent Defection Mutual Cooperation Coordination Failure 34

35 Network Distribution for UB=30 Sampled at End of Generation Network Distribution for HighT: Number of Runs Network Distance Mutual Cooperation Coordination Failure 35

36 Network Distribution for UB=30 Sampled at End of Generation Network Distribution for HighT: Number of Runs Network Distance Mutual Cooperation Coordination Failure 36

37 Four Main Strands of ACE Research Normative Understanding (institutional design,, policy selection, ) Empirical Understanding (possible reasons for empirical regularities) Qualitative Insight/Theory Generation (self-organization of decentralized markets, ) Methodological Advancement (representation, visualization, empirical validation, ) 37

38 ACE and Institutional Design Key Issue: Does an institutional design ensure efficient, fair, and orderly social outcomes over time despite attempts by participants to game the design for their own personal advantage? ACE Approach: Construct an agent-based world capturing salient aspects of the institutional design. Introduce agents with behavioral dispositions, needs, goals, beliefs, etc. Let the world evolve. Observe and evaluate resulting social outcomes. EXAMPLES: Unemployment benefit programs, Internet auctions, stock markets, negotiation protocols, electricity markets 38

39 ACE and Empirical Regularities Key Issue: Is there a causal explanation for persistently observed empirical regularities? ACE Approach: Construct an agent-based world capturing salient aspects of the empirical situation. Investigate whether the empirical regularities can be reliably generated as outcomes in this world. Example: ACE financial market research seeking the simultaneous explanation of financial market stylized facts afinance.htm 39

40 ACE and Qualitative Analysis Illustrative Issue: What are the performance capabilities of decentralized markets? (Adam Smith, F. von Hayek, John Maynard Keynes, J. Schumpeter...) ACE Approach: Construct an agent-based world qualitatively capturing key aspects of decentralized market economies (firms, consumers, circular flow, limited information, ) Introduce traders with behavioral dispositions, needs, goals, beliefs, etc. Let the world evolve. Observe the degree of coordination that results. EXAMPLES: Decentralized exchange economies (no Walrasian Auctioneer ), double-auction markets (learning traders vs. zero intelligence traders), 40

41 Potential Disadvantages of ACE for Economic Modeling Intensive experimentation is often needed (fine sweeps of parameter ranges to attain robust findings) Multi-peaked rather than central-tendency outcome distributions can arise (strong path dependence possible) Can be difficult to ensure platform robustness (i.e., results that are independent of the hardware and/or software implementation of a model) Effort to gain computer modeling skills can be significant (creative computer modeling as opposed to use of existing comp labs requires good programming knowledge) 41

42 Potential Advantages of ACE for Economic Modeling Permits systematic experimental study of empirical regularities, economic institutions, and dynamic behaviors of complex economic processes in general. Facilitates creative experimentation with realistically rendered economic processes: - Using ACE comp labs, researchers/students can evaluate interesting conjectures of their own devising, with immediate feedback and no original programming required - Modular form of ACE software permits relatively easy modification/extension of features. 42

43 ACE Resources ACE Website ACE Handbook (Tesfatsion & Judd, Handbooks in Economics Series, North-Holland, 2006, 904pp) 43

44 44

45 Current ACE Research Areas ( ) Learning and embodied cognition Network formation Evolution of norms Specific market case studies (labor, electricity, finance ) Industrial organisation Technological change and growth Multiple-market economies Market design Automated markets and software agents Development of computational laboratories Parallel experiments (real and computational agents) Empirical validation. and many more areas as well! 45